Semi-supervised Learning Research Articles

Triplet Adaptation Framework for Robust Semi-Supervised Learning.

Semi-supervised learning (SSL) suffers from severe performance degradation when labeled and unlabeled data come from inconsistent and imbalanced distribution. Nonetheless, there is a lack of theoretical guidance regarding a remedy for this issue. To bridge the gap between theoretical insights and practical solutions, we embark to an analysis of generalization bound of classic SSL algorithms. This analysis reveals that distribution inconsistency between unlabeled and labeled data can cause a significant generalization error bound. Motivated by this theoretical insight, we present a Triplet Adaptation Framework (TAF) to reduce the distribution divergence and improve the generalization of SSL models. TAF comprises three adapters: Balanced Residual Adapter, aiming to map the class distribution of labeled and unlabeled data to a uniform distribution for reducing class distribution divergence; Representation Adapter, aiming to map the representation distribution of unlabeled data to labeled one for reducing representation distribution divergence; and Pseudo-Label Adapter, aiming to align the predicted pseudo-labels with the class distribution of unlabeled data, thereby preventing erroneous pseudo-labels from exacerbating representation divergence. These three adapters collaborate synergistically to reduce the generalization bound, ultimately achieving a more robust and generalizable SSL model. Extensive experiments across various robust SSL scenarios validate the efficacy of our method.

Applications of cluster-based transfer learning in image and localization tasks

Transfer learning can address the issue of insufficient labels in machine learning. Using knowledge in a labeled domain (source domain) can assist in acquiring and learning knowledge in a domain (target domain) that lacks some or all labels. In this paper, we propose a new cluster-based semi-supervised transfer learning (CBSSTL) under a new assumption that samples in the target domain are unlabeled but contain cluster information. Furthermore, we propose a new transfer learning framework and a method for fine-tuning parameters. We tested and compared the proposed method with other unsupervised and semi-supervised transfer learning methods on well-known image datasets. The experimental results demonstrate the effectiveness of the proposed method. Additionally, we created a localization dataset for transfer learning. Finally, we tested and analyzed the proposed method on this dataset. Its particularly challenging nature makes it difficult for our method to work effectively.

Robust Semi-Supervised Learning by Wisely Leveraging Open-Set Data.

Open-set Semi-supervised Learning (OSSL) holds a realistic setting that unlabeled data may come from classes unseen in the labeled set, i.e., out-of-distribution (OOD) data, which could cause performance degradation in conventional SSL models. To handle this issue, except for the traditional in-distribution (ID) classifier, some existing OSSL approaches employ an extra OOD detection module to avoid the potential negative impact of the OOD data. Nevertheless, these approaches typically employ the entire set of open-set data during their training process, which may contain data unfriendly to the OSSL task that can negatively influence the model performance. This inspires us to develop a robust open-set data selection strategy for OSSL. Through a theoretical understanding from the perspective of learning theory, we propose Wise Open-set Semi-supervised Learning (WiseOpen), a generic OSSL framework that selectively leverages the open-set data for training the model. By applying a gradient-variance-based selection mechanism, WiseOpen exploits a friendly subset instead of the whole open-set dataset to enhance the model's capability of ID classification. Moreover, to reduce the computational expense, we also propose two practical variants of WiseOpen by adopting low-frequency update and loss-based selection respectively. Extensive experiments demonstrate the effectiveness of WiseOpen in comparison with the state-of-the-art.

An Ensemble-Based Semi-Supervised Learning Approach for Non-Stationary Imbalanced Data Streams with Label Scarcity

Addressing the challenges of learning from multi-class imbalanced data streams, particularly in scenarios with scarce labeled data and concept drift, remains an open problem in the field of data stream mining. Despite the prevalence of such data scenarios in real-world applications, existing approaches have yet to provide effective solutions. In this paper, we propose a novel chunk-based semi-supervised framework, GMCSSEL, which leverages an ensemble of base classifiers trained on micro-cluster centers with labels inferred through a graph-fusion and label propagation process. Our approach incorporates chunk-based incremental label propagation by integrating both current and previously inferred label information into the propagation equation, with regularization parameters applied to control their influence. Furthermore, our method includes a novel concept drift detection mechanism specifically designed for imbalanced data with label scarcity. The imbalanced data problem is addressed through a combination of graph fusion, label matrix normalization, and SMOTE techniques. Experimental results on synthetic data streams with varying class ratios and concept drifts, as well as real multi-class streams, demonstrate the superior classification performance of our approach compared to the IOE semi-supervised algorithm. Our method achieves an average increase in evaluation metrics of 12.5% across multiple data streams, with improvements ranging from 8% to 18% in G-mean, AUC, Kappa, and F1-score metrics. Statistical analysis confirms that these improvements are significant, highlighting the robustness of our approach in handling non-stationary imbalanced data streams with label scarcity.

IGCN: A Provably Informative GCN Embedding for Semi-Supervised Learning With Extremely Limited Labels.

Graph Neural Networks (GNNs) have gained much more attention in the representation learning for the graph-structured data. However, the labels are always limited in the graph, which easily leads to the overfitting problem and causes the poor performance. To solve this problem, we propose a new framework called IGCN, short for Informative Graph Convolutional Network, where the objective of IGCN is designed to obtain the informative embeddings via discarding the task-irrelevant information of the graph data based on the mutual information. As the mutual information for irregular data is intractable to compute, our framework is optimized via a surrogate objective, where two terms are derived to approximate the original objective. For the former term, it demonstrates that the mutual information between the learned embeddings and the ground truth should be high, where we utilize the semi-supervised classification loss and the prototype based supervised contrastive learning loss for optimizing it. For the latter term, it requires that the mutual information between the learned node embeddings and the initial embeddings should be high and we propose to minimize the reconstruction loss between them to achieve the goal of maximizing the latter term from the feature level and the layer level, which contains the graph encoder-decoder module and a novel architecture GCN Info. Moreover, we provably show that the designed GCN Info can better alleviate the information loss and preserve as much useful information of the initial embeddings as possible. Experimental results show that the IGCN outperforms the state-of-the-art methods on 7 popular datasets.

Searching to Exploit Memorization Effect in Deep Learning With Noisy Labels.

Sample selection approaches are popular in robust learning from noisy labels. However, how to control the selection process properly so that deep networks can benefit from the memorization effect is a hard problem. In this paper, motivated by the success of automated machine learning (AutoML), we propose to control the selection process by bi-level optimization. Specifically, we parameterize the selection process by exploiting the general patterns of the memorization effect in the upper-level, and then update these parameters using predicting accuracy obtained from model training in the lower-level. We further introduce semi-supervised learning algorithms to utiilize noisy-labeled data as unlabeled data. To solve the bi-level optimization problem efficiently, we consider more information from the validation curvature by the Newton method and cubic regularization method. We provide convergence analysis for both optimization methods. Results show that while both methods can converge to an (approximately) stationary point, the cubic regularization method can find better local optimal than the Newton method with less time. Experiments on both benchmark and real-world data sets demonstrate that the proposed searching method can lead to significant improvements upon existing methods. Compared with existing AutoML approaches, our method is much more efficient on finding a good selection schedule.

Multi-Level Cross-Modal Interactive-Network-Based Semi-Supervised Multi-Modal Ship Classification

Ship image classification identifies the type of ships in an input image, which plays a significant role in the marine field. To enhance the ship classification performance, various research focuses on studying multi-modal ship classification, which aims at combining the advantages of visible images and infrared images to capture complementary information. However, the current methods simply concatenate features of different modalities to learn complementary information, which neglects the multi-level correlation between different modalities. Moreover, the existing methods require a large amount of labeled ship images to train the model. How to capture the multi-level cross-modal correlation between unlabeled and labeled data is still a challenge. In this paper, a novel semi-supervised multi-modal ship classification approach is proposed to solve these issues, which consists of two components, i.e., multi-level cross-modal interactive network and semi-supervised contrastive learning strategy. To learn comprehensive complementary information for classification, the multi-level cross-modal interactive network is designed to build local-level and global-level cross-modal feature correlation. Then, the semi-supervised contrastive learning strategy is employed to drive the optimization of the network with the intra-class consistency constraint based on supervision signals of unlabeled samples and prior label information. Extensive experiments on the public datasets demonstrate that our approach achieves state-of-the-art semi-supervised classification effectiveness.

Role of Machine and Deep Learning Algorithms in Secure Intrusion Detection Systems (IDS) for IOT & Smart Cities

In this study the authors have examines various machine learning algorithms that could be used in IDS for making secure IoT and Smart Cities. The study examines various deep learning architectures of supervised, unsupervised, and semi-supervised learning methods to improve security and resource usage. Federated learning, edge computing, explainable AI, adversarial machine learning defense, and transfer learning are also explored for smart farming and IoT challenges. Machine learning has the potential to improve security and agricultural sustainability, but it must be researched and developed.

Toward Industrial Densely Packed Object Detection: A Federated Semi-Supervised Learning Approach

Toward Industrial Densely Packed Object Detection: A Federated Semi-Supervised Learning Approach

OPN: Open-Set Semi-Supervised Learning for Intelligent Fault Diagnosis of Rotating Machinery

OPN: Open-Set Semi-Supervised Learning for Intelligent Fault Diagnosis of Rotating Machinery

Physics-informed two-tier neural network for non-linear model order reduction

AbstractIn recent years, machine learning (ML) has had a great impact in the area of non-intrusive, non-linear model order reduction (MOR). However, the offline training phase often still entails high computational costs since it requires numerous, expensive, full-order solutions as the training data. Furthermore, in state-of-the-art methods, neural networks trained by a small amount of training data cannot be expected to generalize sufficiently well, and the training phase often ignores the underlying physical information when it is applied with MOR. Moreover, state-of-the-art MOR techniques that ensure an efficient online stage, such as hyper reduction techniques, are either intrusive or entail high offline computational costs. To resolve these challenges, inspired by recent developments in physics-informed and physics-reinforced neural networks, we propose a non-intrusive, physics-informed, two-tier deep network (TTDN) method. The proposed network, in which the first tier achieves the regression of the unknown quantity of interest and the second tier rebuilds the physical constitutive law between the unknown quantities of interest and derived quantities, is trained using pretraining and semi-supervised learning strategies. To illustrate the efficiency of the proposed approach, we perform numerical experiments on challenging non-linear and non-affine problems, including multi-scale mechanics problems.

Advancing Additive Manufacturing Through Machine Learning Techniques: A State-of-the-Art Review

In the fourth industrial revolution, artificial intelligence and machine learning (ML) have increasingly been applied to manufacturing, particularly additive manufacturing (AM), to enhance processes and production. This study provides a comprehensive review of the state-of-the-art achievements in this domain, highlighting not only the widely discussed supervised learning but also the emerging applications of semi-supervised learning and reinforcement learning. These advanced ML techniques have recently gained significant attention for their potential to further optimize and automate AM processes. The review aims to offer insights into various ML technologies employed in current research projects and to promote the diverse applications of ML in AM. By exploring the latest advancements and trends, this study seeks to foster a deeper understanding of ML’s transformative role in AM, paving the way for future innovations and improvements in manufacturing practices.

Local field potential-based brain-machine interface to inhibit epileptic seizures by spinal cord electrical stimulation.

This study proposes a closed-loop brain-machine interface (BMI) based on spinal cord stimulation to inhibit epileptic seizures, applying a semi-supervised machine learning approach that learns from Local Field Potential (LFP) patterns acquired on the pre-ictal (preceding the seizure) condition.
 
Approach. LFP epochs from the hippocampus and motor cortex are band-pass filtered from 1 to 13 Hz, to obtain the time-frequency representation using the continuous Wavelet transform, and successively calculate the phase lock values (PLV). As a novelty, the Z-score-based PLV normalization using both modified k-means and Davies-Bouldin's measure for clustering is proposed here. Consequently, a generic seizure's detector is calibrated for detecting seizures on the normalized PLV, and enables the spinal cord stimulation for periods of 30 s in a closed-loop, while the BMI system detects seizure events. To calibrate the proposed BMI, a dataset with LFP signals recorded on five Wistar rats during basal state and epileptic crisis was used. The epileptic crisis was induced by injecting pentylenetetrazol (PTZ). Afterwards, two experiments without/with our BMI were carried out, inducing epileptic crisis by PTZ in Wistar rats. 

Main Results. Stronger seizure events of high LFP amplitudes and long time periods were observed in the rat, when the BMI system was not used. In contrast, short-time seizure events of relative low intensity were observed in the rat, using the proposed BMI. The proposed system detected on unseen data the synchronized seizure activity in the hippocampus and motor cortex, provided stimulation appropriately, and consequently decreased seizure symptoms. 

Significance. Low-frequency LFP signals from the hippocampus and motor cortex, and cord spinal stimulation can be used to develop accurate closed-loop BMIs for early epileptic seizures inhibition, as an alternative treatment.

Almost exact recovery in noisy semi-supervised learning

Abstract Graph-based semi-supervised learning methods combine the graph structure and labeled data to classify unlabeled data. In this work, we study the effect of a noisy oracle on classification. In particular, we derive the maximum a posteriori (MAP) estimator for clustering a degree corrected stochastic block model when a noisy oracle reveals a fraction of the labels. We then propose an algorithm derived from a continuous relaxation of the MAP, and we establish its consistency. Numerical experiments show that our approach achieves promising performance on synthetic and real data sets, even in the case of very noisy labeled data.

SSML: Semi-supervised metric learning with hard samples for hyperspectral image classification

Deep learning is widely used in hyperspectral image (HSI) classification due to its powerful learning capabilities. However, its excellent performance typically requires a large number of samples, which can be time-consuming and labor-intensive to produce. The limitation of available samples greatly constrains the model's generalization performance. To alleviate the sample pressure, we propose a semi-supervised metric learning method for HSI classification that focuses on hard samples and can obtain reliable pseudo-labels through multiscale prediction. Additionally, we employ a hard sample learning strategy for network training, which concentrates on enhancing network discrimination by adaptively optimizing intra-class and inter-class distances. Performance tests on three public datasets indicate that the proposed method surpasses other state-of-the-art methods across multiple validation metrics.

AI-driven automatic generation and rendering of game characters

Abstract. This paper provides a comprehensive review of AI-driven techniques for the automatic generation and rendering of game characters, with a particular focus on Generative Adversarial Networks (GANs) and Variational Autoencoders (VAEs). GANs, using adversarial training approaches, have revolutionized character development by producing incredibly realistic and aesthetically pleasing gaming characters. VAEs, despite frequently encountering issues like image blurriness, provide an alternate strategy that emphasizes diversity and originality in the generated content. Additionally, conditional models that enable more individualized and regulated character generation are explored, as well as hybrid models that combine the best features of both GANs and VAEs. The difficulties with mode collapse in GANs and the requirement for big datasets for both GANs and VAEs are also covered, along with some possible fixes like transfer learning and semi-supervised learning strategies. This analysis emphasizes the growing significance of AI-driven game character generation in the gaming industry by highlighting its current state, problems, and future directions.

Automatic delineation of cervical cancer target volumes in small samples based on multi-decoder and semi-supervised learning and clinical application

Radiotherapy has been demonstrated to be one of the most significant treatments for cervical cancer, during which accurate and efficient delineation of target volumes is critical. To alleviate the data demand of deep learning and promote the establishment and promotion of auto-segmentation models in small and medium-sized oncology departments and single centres, we proposed an auto-segmentation algorithm to determine the cervical cancer target volume in small samples based on multi-decoder and semi-supervised learning (MDSSL), and we evaluated the accuracy via an independent test cohort. In this study, we retrospectively collected computed tomography (CT) datasets from 71 pelvic cervical cancer patients, and a 3:4 ratio was used for the training and testing sets. The clinical target volumes (CTVs) of the primary tumour area (CTV1) and pelvic lymph drainage area (CTV2) were delineated. For definitive radiotherapy (dRT), the primary gross target volume (GTVp) was simultaneously delineated. According to the data characteristics for small samples, the MDSSL network structure based on 3D U-Net was established to train the model by combining clinical anatomical information, which was compared with other segmentation methods, including supervised learning (SL) and transfer learning (TL). The dice similarity coefficient (DSC), 95% Hausdorff distance (HD95) and average surface distance (ASD) were used to evaluate the segmentation performance. The ability of the segmentation algorithm to improve the efficiency of online adaptive radiation therapy (ART) was assessed via geometric indicators and a subjective evaluation of radiation oncologists (ROs) in prospective clinical applications. Compared with the SL model and TL model, the proposed MDSSL model displayed the best DSC, HD95 and ASD overall, especially for the GTVp of dRT. We calculated the above geometric indicators in the range of the ground truth (head-foot direction). In the test set, the DSC, HD95 and ASD of the MDSSL model were 0.80/5.85 mm/0.95 mm for CTV1 of post-operative radiotherapy (pRT), 0.84/ 4.88 mm/0.73 mm for CTV2 of pRT, 0.84/6.58 mm/0.89 mm for GTVp of dRT, 0.85/5.36 mm/1.35 mm for CTV1 of dRT, and 0.84/4.09 mm/0.73 mm for CTV2 of dRT, respectively. In a prospective clinical study of online ART, the target volume modification time (MTime) was 3–5 min for dRT and 2–4 min for pRT, and the main duration of CTV1 modification was approximately 2 min. The introduction of the MDSSL method successfully improved the accuracy of auto-segmentation for the cervical cancer target volume in small samples, showed good consistency with RO delineation and satisfied clinical requirements. In this prospective online ART study, the application of the segmentation model was demonstrated to be useful for reducing the target volume delineation time and improving the efficiency of the online ART workflow, which can contribute to the development and promotion of cervical cancer online ART.

A novel method for assessment rooftop PV potential based on remote sensing images

The assessment of rooftop photovoltaic (PV) potential is highly significant for energy policy formulation. With the rapid development of computer vision (CV) and remote sensing imagery, utilizing CV to extract rooftop information is an ideal approach. However, deep learning requires a large amount of accurately annotated data, and annotating remote sensing images is a labor-intensive task. This limitation hinders the application of deep learning in rooftop PV potential assessment. To address this issue, this paper proposes a semi-supervised learning (SSL)-based segmentation model to extract rooftop information from remote sensing images. Subsequently, a rooftop classification method is proposed to categorize rooftops into several classes and estimate their rooftop PV available area ratios. Finally, the total available rooftop PV area in urban areas is evaluated, and the potential rooftop PV installed capacity and power generation are calculated. This method is applied in the Longhu District of Shantou City, Guangdong Province. The evaluation results show that the total rooftop area in Longhu District is 17.2 km2, with a rooftop PV available area of 12.7 km2. It is estimated that the rooftop PV installed capacity in Longhu District is 1849.4 MW, with an annual power generation of 2219.3 GWh.

Developing a 10-Layer Retinal Segmentation for MacTel Using Semi-Supervised Learning.

Automated segmentation software in optical coherence tomography (OCT) devices is usually developed for and primarily tested on common diseases. Therefore segmentation accuracy of automated software can be limited in eyes with rare pathologies. We sought to develop a semisupervised deep learning segmentation model that segments 10 retinal layers and four retinal features in eyes with Macular Telangiectasia Type II (MacTel) using a small labeled dataset by leveraging unlabeled images. We compared our model against popular supervised and semisupervised models, as well as conducted ablation studies on the model itself. Our model significantly outperformed all other models in terms of intersection over union on the 10 retinal layers and two retinal features in the test dataset. For the remaining two features, the pre-retinal space above the internal limiting membrane and the background below the retinal pigment epithelium, all of the models performed similarly. Furthermore, we showed that using more unlabeled images improved the performance of our semisupervised model. Our model improves segmentation performance over supervised models by leveraging unlabeled data. This approach has the potential to improve segmentation performance for other diseases, where labeled data is limited but unlabeled data abundant. Improving automated segmentation of MacTel pathology on OCT imaging by leveraging unlabeled data may enable more accurate assessment of disease progression, and this approach may be useful for improving feature identification and location on OCT in other rare diseases as well.

Physics-Informed Data-Driven Approaches to State of Health Prediction of Maritime Battery Systems

Battery systems are increasingly being used for powering ocean going ships, and the number of fully electric or hybrid ships relying on battery power for propulsion and maneuvering is growing. In order to ensure the safety of such electric ships, it is of paramount importance to monitor the available energy that can be stored in the batteries, and classification societies typically require that the state of health (SOH) of the batteries can be verified by independent tests – annual capacity tests. However, this paper discusses physics-informed data-driven approaches to online diagnostics for state of health monitoring of maritime battery systems based on a combination of physical knowledge, physic-based models, insights from extensive characterization tests and operational sensor data collected from the batteries during actual operation. This represents an alternative approach to the annual capacity tests for electric ships that is found to be sufficiently robust and accurate under certain conditions. Previous attempts with purely data-driven models, including both cumulative and snapshot models, semi-supervised learning and simple models based on the state of charge did not achieve the required reliability and accuracy for them to be utilized in a ship classification perspective, as presented at previous PHM conferences. However, preliminary results from the physics-informed data-driven method presented in this paper indicate that it can be relied on for independent verification of state of health as an alternative to physical tests. It has been tested on battery cells cycled in laboratory degradation tests as well as on field returns from batteries onboard ships in service. Notwithstanding, further validation and verification of the method is recommended to further build confidence in the model predictions.